THE PENNSYLVANIA STATE UNIVERSITY SCHREYER HONORS COLLEGE DEPARTMENT OF BIOCHEMISTRY AND MOLECULAR BIOLOGY

Transcription

1 THE PENNSYLVANIA STATE UNIVERSITY SCHREYER HONORS COLLEGE DEPARTMENT OF BIOCHEMISTRY AND MOLECULAR BIOLOGY GENOME-WIDE MICROARRAY ANALYSIS IN A CASE-CONTROL STUDY REVEALS EVELATED LEVEL OF GLOBAL COPY NUMBER BURDEN IN AUTISM KIAN HUI YEOH SPRING 2012 A thesis submitted in partial fulfillment of the requirements for a baccalaureate degree in BIOTECHNOLOGY with honors in BIOTECHNOLOGY Reviewed and approved* by the following: Maria Krasilnikova Research Assistant Professor Department of Biochemistry and Molecular Biology Thesis Supervisor and Research Advisor Scott B. Selleck Professor and Department Head Department of Biochemistry and Molecular Biology Thesis Supervisor and Research Advisor David S. Gilmour Professor of Molecular and Cell Biology Department of Biochemistry and Molecular Biology Honors Adviser Wendy Hanna-Rose Associate Professor of Biochemistry and Molecular Biology Associate Department Head for Undergraduate Studies * Signatures are on file in the Schreyer Honors College.

2 ABSTRACT Copy number variations (CNV), or structural genomic variations, have recently been shown to be implicated in and associated with numerous human neurodevelopmental disorders such as autism spectrum disorders (ASD), schizophrenia, epilepsy, mental retardations, developmental delays, bipolar disorder, intellectual disability (ID) and many other human diseases. In this study, we used a custom-design DNA microarray targeted at regions flanked by segmental duplications (SD) to evaluate and compare global copy number burden between 274 autistic (AU) patients and 280 typically developing (TD) individuals from the general population. We found that there was an elevated level of global CNV burden in AU individuals (Mann-Whitney Test, U= 23449, p= ). Additionally, we showed that total average duplication length in AU individuals was also significantly higher than that of controls (Mann- Whitney Test, U= 28129, p= ) compared to total average deletion length (Mann-Whitney Test, U= 20804, p=0.432). Interestingly, this differential level of global burden between cases and controls was mostly detected in the genomic backbone or non-hotspot region (Mann- Whitney Test, U=7992, p=0.0128), but not the genomic hotspot regions (Mann-Whitney Test, U=28987, p=0.513) as we originally expected. The data from this case-control study demonstrated an elevated level of copy number changes in autistic individuals, which was primarily represented by large duplication events located in the non-hotspot or genomic backbone regions. The findings from this study also implicated that there may be an alternative mechanism other than non-allelic homologous recombination (NAHR) involved in this significant copy number variation in autism. Page i

3 TABLE OF CONTENTS List of figures and table.. iii Acknowledgements.. iv Introduction. 1 Genomic structural variants and human disorders. 1 Detecting CNVs using microarray-based method... 1 Association analysis of variants and neurodevelopmental disorders.. 4 Overview of methodology... 4 Materials and Methods 6 Ethical Considerations 6 Study Cohort and DNA Samples 6 Design of the custom oligoarray for acgh experiment. 8 Samples labeling, microarray hybridization and scanning 14 Array analysis and CNV callings 14 Statistical analysis of CNV burdens 17 Results. 19 Qualification control of samples for CNV callings. 19 Elevated level of global CNV burden in autistic individuals.. 20 Elevated level of CNV burden is largely found in genomic backbone Elevated level of non-rare and nonpathogenic copy number burden.. 26 Large events constitute the elevated level of duplication 29 Discussion Autistic individuals have higher level of global CNV burden 32 Increased level of CNV burden in autism is manifested by large duplications.. 33 Copy number burden is elevated in non-hotspot genomic regions. 34 Genomic instability, copy number variations and autism Limitations of the Hotspot v1.0 array. 36 Future directions and implications of this study. 37 References Academic Vita. 49 Page ii

4 LIST OF FIGURES AND TABLES 1. Figure 1A: Schematic representation of acgh experimental procedure Figure 1B: Graph output example produced from UCSC genome browser Figure 2A: The design of the HS1 array Figure 2B: Scatter plot of probes distribution Figure 3: Schematic overview of our research pipeline Figure 4: CNV burden comparison in whole genomic regions Figure 5: CNV burden comparison in genomic non-hotspot regions Figure 6: CNV burden comparison in genomic hotspot regions Figure 7: CNV burden comparison in genomic hotspot and hotspot-associated regions Figure 8: Non-pathogenic CNV burden comparisons in whole genomic regions Figure 9: Frequency distribution plots of CNV events in both cohorts Table 1: Summary of the CHARGE study cohort Table 2: The 107 hotspot chromosomal regions Table 3: Summary of the number of samples that passed QC Table 4: CNV burden lengths (duplications and deletions) in each AU and TD cohort Table 5: List of rare pathogenic variants removed from the HMM calls 28 Page iii

5 ACKNOWLEDGEMENTS I would like to express my utmost gratitude to my research advisor and thesis supervisor, Dr. Maria Krasilnikova, for her constant, devoted guidance and assistance throughout the last two years I have been working in the lab. Without her, my academic and research achievement would not have been where it is now. She has been a very dedicated and caring professor, and has been very helpful in conducting my own independent research. I would also like to specially thank Dr. Scott B. Selleck, Professor and Department Head of Biochemistry and Molecular Biology, who is also my research and thesis supervisor, for the precious opportunity he has given me to participate in a research as a student intern in Dr. Evan Eichler s lab at University of Washington, Seattle, over the summer in I have certainly gained a lot of invaluable experience and advanced research techniques, as an undergraduate student, from the research collaboration I have been involving in. In addition to that, I truly appreciate Dr. Santhosh Girirajan, who is an outstanding senior investigator and my summer research supervisor, and Carl Baker, who is an amazing research scientist, for their guidance and teaching during my summer internship in Eichler Lab. Last but not least, I would like to thank all my fellow lab colleagues, Qiao Kai Law, Su Jen Khoo, Abhinaya Srikanth, Nari Kim, and Stephen Wellard for all the support and assistance they have given to me over the course of my research in the lab. Page iv

6 INTRODUCTION Genomic structural variants and human disorders Structural genomic variations or rearrangements are changes in chromosomal structure that could possibly lead to inversions, translocations, duplications and deletions of any parts or sections of the chromosome, which could be as small as one single nucleotide or as large as one whole chromosome. The most pronounced types of genomic variations are duplications and deletions that cumulatively constitute copy number variation (CNV), which results in an atypical number of copies of one or more segments in the human genome. CNV contributes to an estimated portion of 12-15% of the human genome, with each variation ranging from a few thousand base pairs to several million base pairs of DNA, significantly contributing to human genetic heterogeneity [1]. Studies emerging in recent years have shown that CNV was implicated in a wide range of neurodevelopmental disorders, including but not limited to intellectual disability (ID) [2,3], autism [4,5], bipolar disorder [6], epilepsy [7], schizophrenia [8], as well as attention deficit hyperactivity disorder (ADHD) [9]. In addition to that, a study has shown that an increased CNV burden had a positive correlation with disease severity [10], that is, a larger CNV always leads to more severe neurodevelopmental disorders. Detecting CNVs using microarray-based method The two most commonly used methods in detecting CNV are SNP-based microarray and arraycomparative genomic hybridization (acgh), even though recent advances in DNA sequencing technology have enabled the use of next-generation whole-genome sequencing, providing a higher accuracy of CNV detections [11]. Both SNP-based microarray and acgh detect CNV through the comparison of test samples to a reference sample, which is then analyzed by a Page 1

7 computer for producing specific output signals. These signals are then normalized using certain mathematical or statistical algorithms to determine the output signal intensity: increased intensity of the output signal reflects a duplication, which is literally a presence of two or more copies of a particular segment of the genome, whereas a decreased intensity reflects deletions. Figure 1A illustrates the general procedure of acgh, and Figure 1B shows an example of its generated graph output. Figure 1. A) Schematic representation of acgh experimental procedure: DNA samples from both test case and reference control are each labeled with a different fluorescent dye,. After mixing, the labeled DNA samples are then hybridized to an array slide, which has specific probes covering genomic regions attached to its surface. After hybridization, the array is then subject to the computer scanning to produce a signal output (not shown). B) Graph output example produced from UCSC genome browser: Graphs are plotted and colored based on a specific threshold of the log intensity ratio; Red: deletion; Green: duplication; Black: the log intensity ratio falls below the threshold of 1.5. Horizontal axis represents the location across the genome. (Figures Courtesy of Dr. Santhosh Girirajan and Dr. Scott Selleck) Page 2

8 Page 3

9 Association analysis of variants and neurodevelopmental disorders There are several approaches to studying how CNV could contribute to pathogenicity in neurodevelopmental disorders. One of those is to search for rare and pathogenic CNVs that occur in affected individuals but not in unaffected controls for each of these disorders using whole genome array studies. These variants are usually overlapping with one or more important and critical genes, where a change in a copy number could potentially result in phenotypic and functional alterations leading to the pathogenesis of a disease. Through this approach of association analysis of rare and functional genic variants, several studies have been able to associate one or more CNVs with a particular disorder. For instance, 15q duplication is associated with Angelman/Prader-Willi syndromes and autism spectrum disorders [12], 7q11.23 deletion causes Williams-Beuren syndrome [13-15] whereas its reciprocal duplication is associated with autism, language or speech delay, and mental retardation [16,17]; 16p12.1 microdeletion is associated with a severe developmental delay [18], and 17q12 duplication has associated risks of intellectual disability, seizures [19] and autism [20]. However, it is important to note that although numerous studies have implicated the association and presence of these CNVs with certain disease phenotypes, these variants are not necessarily responsible for the direct causal mechanism of the disease. Hence, it is crucial to fully understand the underlying mechanism by which CNV causes diseases. Overview of methodology Whereas numerous studies have associated several rare and pathogenic CNVs with a range of neurobehavioral disorders such as autism, we took an intuitive approach to analyze the relationship between CNV burden and disease phenotypic susceptibility, specifically autism, to Page 4

10 help us understand the relevance of copy number instability across a wide range of these disorders. We performed a systematic analysis on 274 autistic (AU) individuals and 280 typically developing (TD) children from the general population to address the relative contribution of CNV in autism. We hypothesized that there is a greater level of CNV burden in autistic cases compared to controls. To evaluate this, a whole-genome custom microarray was designed to target genomic hotspots for the CGH to detect and identify CNVs in both cases and controls. The data obtained from this microarray was statistically analyzed to assess whether there is an elevated level of copy number burden in AU individuals. Page 5

11 MATERIALS AND METHODS Ethical Considerations Patients from each study cohort, including both cases and controls, were recruited in accordance with appropriate human subjects approval and informed consent. Informed consent was also acquired from the patients to obtain DNA samples in conjunction with IRB protocols and guidelines. Study Cohort and DNA Samples A total of 553 DNA samples were acquired from the Childhood Autism Risks from Genetics and Environment (CHARGE) Study[21], which was conducted by the Medical Investigations of Neurodevelopmental Disorders (MIND) Institute at University of California, Davis. These DNA samples were obtained from whole blood of the patients who were selected based on complete criteria for Autistic Disorder (OMIM ) using ADOS[22], and the Autism Diagnostic Interview, Revised (ADI-R)[23]. On the other hand, individuals with significant impairments in visual, audio, or motor skills, serious birth complications such as extended NICU stay, diagnosis of syndromic autism and known genetic causes of autism including Fragile X Syndrome, were excluded. Evaluation of diagnosis, assessment of cognitive performance, and characterization of phenotypes of each individual as well as data entry and collection were performed at the MIND institute. Aside from the AU individuals, the study cohort also included typically developing (TD) children without any official diagnosis of autism. However, it is worth noting that the TD individuals group obtained from the general population included a small number of individuals with mental retardation (MR)/developmental delay (DD) despite the fact that they had an atypical cognitive performance or other phenotypic considerations non-related to autism. In Page 6

12 summary, the CHARGE study included a total of 273 AU individuals with the following ethnicity breakdown: Caucasian (143), Hispanic (74), Mixed Race (30), Asian (20), and African American (6). A control group consisted of 280 TD individuals with the ethnicities of Caucasian (147), Hispanic (83), Mixed Race (37), Asian (7), African American (5), and Pacific Islander/Hawaii Native (1). Table 1 summarizes the study cohort of both AU and TD individuals, with proportion of males shown in percentage for each group. Table 1. Summary of the CHARGE study cohort with details of ethnicity breakdown. TD: Typically developing individuals; AU: autistic individuals. % male reflects the percentage of male individuals in each study cohort. Ethnicity TD (%male) AU (% male) Total Samples Hispanic 83 (0.81) 74 (0.83) 157 White 147 (0.83) 143 (0.88) 290 Asian 7 (0.66) 20 (0.90) 27 African American 5 (0.8) 6 (0.83) 11 Mixed 37 (0.58) 30 (0.86) 67 Native Hawaiian 1(1.0) - 1 Total samples Page 7

13 Design of the custom oligoarrays for acgh experiment As mentioned earlier, array comparative genomic hybridization (acgh) was used in this study to detect CNV in the AU and TD DNA samples. To perform acgh on the DNA samples, we used a custom-designed 12-plex oligoarray Hotspot v1.0 (HS1) from Roche Nimblegen Systems, Inc. This array contained a total number of 135,000 probes with two different probe densities coverage: higher density probe coverage in 107 genomic hotspots regions (Table 2) flanked by segmental duplications (SD), and a lower probe density in the genomic backbone that included the rest of the genome. The median probe spacing in the genomic hotspots regions was 2.6 kb, while the genomic backbone region had a mean probe spacing of 36 kb. Although the median density of the chip was designed to be approximately 2.6 kb in the genomic hotspots and 36 kb in genomic backbone, the limitations of the chip design (probe assignment restricted to only up to five mismatches) precluded uniform distribution of the probes throughout the genome. Therefore, the actual probe density varied across regions of the human genome. Figure 2 depicts the distribution of the probe spacing in the Hotspot v1.0 array in details. In addition to Hotspot v1.0 array, another custom targeted 2 400K Agilent chip (Agilent Technologies), with median probe spacing of 500 bp in the genomic hotspots and probe spacing of 14 kb in the genomic backbone, were used in several validation experiments to confirm the rare pathogenic events detected using the Hotspot v1.0 array. Page 8

14 Table 2. The description of the 107 hotspot chromosomal regions (only 87 regions are shown here, since some of the hotspot regions were combined) and probes utilized in the Hotspot v1.0 chip. Note that although the median spacing in the hotspot regions was around 2.6 kb, the average distance between probes was different within each particular region. Some of the regions are known for association of several genomic disorders as indicated. (Table courtesy of Dr. Santhosh Girirajan) Chr Start End Size Number of probes Probe spacing Chromosomal segments/regions chr , q34.2 chr , p33 chr , p13.3 chr ,300, q21.1* chr ,537, q21.1* chr , q44 chr ,321, p11.2 chr ,880, q11.1q11.2 chr ,841, q11.2q13 chr ,100, q11.2q13 chr ,265, q21.3 chr , q37.1* chr ,843, p25.3 chr ,083, q21.1q21.3 chr ,003, q29* chr ,393, p16.2p16.1* chr , q13.2 chr , q13.2 chr , q13.3 chr , q26 chr ,358, p15.33 chr ,951, p14.3p14.1 chr ,809, q13.2 chr , q21.1 chr ,997, q35.2q35.3 (Sotos)* chr , q27 chr ,538, p15.1p14.3 chr ,804, p14.2p13 Page 9

15 chr ,629, p13p11.2 chr , q11.21 chr ,786, q11.22 chr ,525, q11.23 (Williams)* chr ,275, q36.1q36.2 chr , p23.3p23.2 chr ,653, p23.1* chr , q24.3* chr ,735, p13.3p13.2 chr ,027, q12q13 chr ,967, q21.32q21.33 chr , q22.1 chr ,642, q22.32 chr , p12.1 chr ,093, q11.22 chr ,989, q23.1* chr ,921, p11.2 chr , q11 chr ,881, q13.2q13.4 chr , p13.31 chr ,029, p11.1q12 chr ,575, q12.11q12.12 chr ,300, q11.2 (BP1-BP2)* chr ,800, q11.2q13.1 (BP2-BP3)* chr ,987, q13.1q13.3 (BP3-BP5)* chr ,685, q24.1q24.2* chr ,225, q24.2q24. 3* chr ,181, q25.2* chr ,573, p13.11p13.3* chr ,300, p12.1p11.2* chr ,454, p11.2* chr ,809, p11.2 chr ,612, q22.1q22.3 chr , p13.3 (Miller-Dieker)* chr ,735, p12 (CMT1A/HNPP)* chr ,750, p11.2p12 (SMS)* chr ,250, p11.2 chr , GAP chr ,155, q12 (RCAD)* chr ,422, q21* chr ,725, q23.1q23.2 chr ,627, p11.2p11.21 chr , p12 chr ,032, q13.12 chr ,234, q13.32q13.33 chr , p13 Page 10

16 chr ,001, q11.2q11.23* chr ,300, q13* chrx , Xp12.1 chrx ,736, Xp11.23p11.22 chrx , Xp11.1 chrx , Xq28 chrx , Xq28 chrx , Xq28 chry ,194, Yp11.2 chry ,139, Yq11.23 chry ,233, Yq11.23q11.2 *Regions associated with known genomic disorders. Page 11

17 Figure 2. A) The design of the HS1 array: The array has a median probe spacing of 2.6 kb in the genomic hotspot regions (regions flanked by SD), and a median probe spacing of 36 kb in other non-hotspot genomic backbone regions. Hotspot CNV: CNVs that are detected within the genomic hotspot region using the HS1 array; hotspot-associated CNV: part of the CNV detected is located in the genomic hotspot; non-hotspot CNV: CNVs detected in the genomic backbone, outside the genomic hotspot regions. B) Scatter plot of probes distribution: This scatter plot shows the size-distribution of CNVs and the density of array probes targeted to the genomic hotspots and non-hotspot regions (see Figure 2A). Note that non-hotspot genomic backbone regions contain two different probe densities (probe spacing of bp and bp, which are labeled (a) and (b)) and the 107 hotspot regions are covered at a probe density ranging from 1000 bp up to 10,000 bp. (Figures courtesy of Dr. Girirajan). Page 12

18 Page 13

19 Samples labeling, microarray hybridization and scanning Labeling and microarray hybridization of DNA samples were performed using the Nimblegen array CGH kit (Roche Nimblegen Systems) according to the manufacturer's instructions, and a previously described protocol [24]. All acgh experiments were performed using a reference DNA sample prepared from whole blood of a single unaffected Caucasian male (Coriell, NA15724 ). DNA test samples, including both AU and TD individuals, were labeled with Cy3 fluorescent dye, while the reference sample was labeled with Cy5 fluorescent dye. Each of the labeled test samples was mixed with the reference sample and hybridized to the custom-designed HS1 12-plex 135K array using a 12 Bay Hybridization System (Roche Nimblegen) according to manufacturer's recommendations. After hybridization and washing, the arrays were scanned using GenePix 4000B Microarray Scanner (Molecular Devices,) according to the manufacturer's instructions. Whereas for the 2 400K Agilent chip array, hybridization experiments were performed as described previously [10] and according to the manufacturer s recommendations. Array analysis and CNV callings Data produced from scanned arrays were extracted, processed and normalized using the NimbleScan software package (Roche Nimblegen). Utilizing the embedded SegMNT algorithm within the software, segments were assigned to the normalized data and log intensity ratio were produced. Log intensity ratios were transformed into z-scores using chromosome-specific means and standard deviations. These z-scores were classified as normal, increased, or decreased copy numbers with the help of HMMSeg algorithm based on a three-state Hidden Markov Model (HMM) [25]. For each sample data, probes were combined into segments if consecutive probes of the same state were less than 50 kb apart; for instance, two probes of the same state (such as Page 14

20 increased in copy number) were merged into a single segment if they were not more than 50 kb from each other. In addition to that, two segments of the same state and the sequences between them would also be called as a single CNV as long as the sequence in between the segments has less than 5 probes and is smaller than 10 kb. For the post-hmm filtering thresholds, we employed stringent quality control (QC) measures to increase the accuracy of our experimental data by reducing false positives. For the first part of the QC measure, which was applied to ensure reliability, we eliminated individual samples that had a standard deviation or Log R ratio larger than These samples were excluded because a large standard deviation was typically associated with the noisiness and inconsistency of the data. Samples that passed the first QC were then subjected to the second part of the QC to eliminate false positive CNV calls in each sample. To be considered a real variant, a CNV had to have an absolute z-score larger than 1.5, and contain 10 or more probes, which are empirically estimated thresholds. Since our array had different probe densities in the hotspot and non-hotspot regions, we separated CNVs into three different groups: hotspot CNV, hotspot-associated CNV, and non-hotspot CNV (Figure 2). Hence, we use different threshold of minimum length or size in calling those CNVs. The minimum length for calling the hotspot CNV and hotspot-associated CNV was 20kb; whereas for calling non-hotspot CNV, the CNV length had to be at least 50kb long. In addition to that, a number of artifacts were also removed from the called CNVs; CNVs were considered artifacts if they were located at the centromere/telomere regions, or were not reliable due to the noisiness of the data for these CNV regions. Through the use of these QC criteria, the HMM data outputs could be filtered and scanned for real variants, which were then collected and subjected to further statistical analysis for CNV burden. Figure 3 summarizes the overview of our research pipeline, from the array preparation and hybridization to statistical analysis. Page 15

21 Figure 3. Schematic overview of our research pipeline. 1) All samples were subjected to HS1 array hybridization, scanning and analysis, producing normalized log intensity ratios as well as z-scores as output. 2) These z-scores were then analyzed by HMMSeg to produce CNV calls as output. 3) QC 1: samples with standard deviation larger than 0.32 were eliminated. 4) QC 2: real variants are required to be equal to or larger than 50kb for the non-hotspot CNV, and 20kb or larger for the hotspot and hotspot-associated CNV. They are also required to contain 10 or more probes, and have an absolute z-score of larger than 1.5. QC 3: CNV calls that are artifacts were removed based on their characteristics (such as location near telomeres or centromeres). Page 16

22 Calling for rare, pathogenic CNVs The detection and calling for rare pathogenic variants were mostly performed using manual curation through utilizing the functionality embedded in the UCSC genome browser. The criteria for calling a rare and potentially pathogenic event were based on the frequency of that particular CNV. In other words, to be considered a rare pathogenic variant, the putative CNV has never been seen in controls or is found in low frequency (10 or less out of 8329 controls), and affects chromosomal regions harboring related functional genes that potentially contribute to the pathogenesis of the disorder. Most of the time, these pathogenic variants had already been reported in many other previous studies. The detected putative pathogenic CNVs were validated using a different platform with relatively much higher probe densities in the array. Statistical analysis of CNV burdens The comparison of CNV burden between AU and TD individuals were done mainly by using two approaches: 1) Comparison of a total average length of CNVs per individual, including duplications and deletions, in autistic and control groups; 2) Comparison of total numbers of CNV per individual, in autistic and control group. This analysis was performed separately for duplications and deletions. CNV length burden was normalized on the total number of kids in each group. We also employed two different approaches to comparing the CNV burdens between AU and TD individuals to investigate whether genomic instability was more pronounced at the hotspot regions or the genomic backbone. First, we performed a global CNV burden analysis in the entire genome, which included the high-density probe regions (the SD or hotspot regions) as well as the low-density probe regions (other unique genomic or non-hotspot regions). Second, Page 17

23 we performed a CNV burden comparison between autistic and controls on the hotspot, hotspotassociated, and non-hotspot regions separately. In addition to analysis of CNV burden on hotspot and non-hotspot regions of the genome, we have also performed copy number burden analysis on both sample cohorts after removing rare, potentially pathogenic CNV calls as detected previously from the HMM calls. Significances of all the differences were tested statistically using Mann-Whitney U-Test. We were not able to use the T-test due to the non-gaussian distribution characteristic of the data sets. A maximum p-value threshold of 0.05 and differential threshold of 5% were expected for the differences to be statistically significant. Page 18

24 RESULTS Qualification control of samples for CNV callings Through utilization of QC parameters mentioned above for calling CNVs in each individual, samples that did not meet the qualification criteria for reliable data, that is, the standard deviation was larger than 0.32, were eliminated. Out of the 553 analyzed samples, 54 did not pass QC. Table 3 below summarizes the ethnicity breakdown of the 499 samples that passed QC. Table 3. Summary of the number of samples that passed QC (having a standard deviation or Log R ratio < 0.32). Out of the initial number of 553 samples, 54 did not pass QC. Ethnicity TD (# of samples not passing QC) AU (# of samples not passing QC) Total Hispanic 81 (2) 57 (17) 138 White 134 (13) 133 (10) 267 Asian African American 4 (1) 6 10 Mixed 30 (7) 27 (3) 57 Native Hawaiian 0 (1) - 0 Total Page 19

25 Elevated level of global CNV burden in autistic individuals Analysis of the CNV burden in the entire genome, including hot spot and non-hot spot regions, showed that the total average CNV burden length per individual was bp in AU patients (total average duplication length: bp; total average deletion length: bp) and bp in TD controls (total average duplication length: bp; total average deletion length: bp) (Table 4). As shown in Figure 4A, the AU patients group showed a significantly increased level of global CNV burden compared to that of TD (Mann-Whitney Test, U= 23449, p= ). The total average duplication length in AU individuals was also significantly higher (Mann-Whitney Test, U= 28129, p= ) (Figure 4B). On the other hand, there was no significant difference in the average total length of deletions between AU and TD individuals (Mann-Whitney Test, U= 20804, p=0.432) (Figure 4B). Table 4. CNV burden lengths (duplications and deletions) in each AU and TD cohort AU individuals TD individuals Average Duplication Length per individual (bp) Average Deletion Length per individual (bp) Total CNV Burden Length per individual (bp) Page 20

26 Figure 4. CNV burden comparison in whole genome regions. A) The AU patients group showed a significantly increased level of global CNV burden compared to that of TD (Mann-Whitney Test, U= 23,449, p= ). B) The total average duplication length in AU individuals was significantly higher (Mann-Whitney Test, U= 28,129, p= ); whereas there was no significant difference in total average deletion length between AU individuals and TD individuals (Mann-Whitney Test, U= 20,804, p=0.432). Page 21

27 Elevated level of CNV burden was largely found in genomic backbone As mentioned earlier, we also performed analysis on the genomic backbone region separately to compare the non-hotspot CNV burden level in AU and TD individuals. The burden comparisons on the genomic backbone shows that the total average CNV burden length per individual was 692,446 bp in AU patients (total average duplication length: 489,378 bp; total average deletion length: 203,068 bp) and 542,985 bp in TD controls (total average duplication length: 372,248 bp; total average deletion length: 170,737 bp). As shown in Figure 5A, the AU patients group showed a significantly increased level of CNV burden compared to that of TD (Mann-Whitney Test, U=7,992, p=0.0128). Similar to the data obtained from the whole genome region analysis, the total average duplication length in AU individuals was significantly higher (Mann-Whitney Test, U=19,986, p=0.0209) (Figure 5B), but there was no significant difference in the comparison of average total length of deletions between AU and TD individual (Mann-Whitney Test, U=2,920.5, p=0.483) (Figure 5B). On the other hand, CNV burden analysis on the hotspot CNVs revealed no significant difference between AU individuals and TD individuals, in terms of total average duplication size (Figure 6B) (Mann-Whitney Test, U=24,022, p=0.944), total average deletion size (Figure 5B) (Mann-Whitney Test, U=24,049, p=0.575), as well as total CNV burden size (Figure 6A) (Mann-Whitney Test, U=28,987, p=0.513). Similar results were also obtained when burden comparison was estimated combining the hotspot CNV and hotspotassociated CNV: there was no statistically significant difference in CNV burden level of AU and TD individuals. The total average duplication size (Figure 7B) (Mann-Whitney Test, U=24,334, p=0.931), total average deletion size (Figure 7B) (Mann-Whitney Test, U=23,045, p=0.830), and total CNV burden size (Figure 7A) (Mann-Whitney Test, U=28,662, p=0.719) were comparable for autistic and control groups. Page 22

28 Figure 5. CNV burden comparison in non-hotspot genomic regions. A) Total average length of CNV burden comparison between AU and TD individuals. The AU patients group showed a significantly increased level of CNV burden compared to that of TD (Mann- Whitney Test, U=?, p=0.0128). B) Total average duplication and deletion lengths comparison between AU and TD individuals. Page 23

29 Figure 6. CNV burden comparison in genomic hotspot regions. A) Total average length of CNV burden comparison between AU and TD individuals. B) Total average duplication and deletion lengths comparison between AU and TD individuals. Page 24

30 Figure 7. CNV burden comparison in genomic hotspot and hotspot-associated regions A) Total average length of CNV burden comparison between AU and TD individuals. B) Total average duplication and deletion lengths comparison between AU and TD individuals. Page 25

31 Elevated level of non-rare and nonpathogenic copy number burden In order to investigate whether the elevated level of copy number burden found in AU individuals was largely due to rare, potentially pathogenic events, we have performed analysis and calculation on the CNV burden after removing and eliminating those functional genic variants (Table 5) from the event calls. This burden comparison was performed on the entire genome regions (including hotspot and non-hotspot regions). Our statistical analysis showed a similar trend to the previously obtained results: the AU cohort showed a significantly increased level of CNV burden compared to that of TD (Mann-Whitney Test, U= , p=0.008) (Figure 8A). Similarly, the total average duplication length in AU individuals was significantly higher (Mann-Whitney Test, U= 21,493, p= 0.022) (Figure 8B), but there was no significant difference in the comparison of average total length of deletions between AU and TD individual (Mann-Whitney Test, U= 20,819.5, p= 0.477) (Figure 8B). Page 26

32 Figure 8. Non-pathogenic CNV burden comparisons for the whole genome regions after rare and potentially pathogenic events were removed. A) Total average length of CNV burden comparison between AU and TD individuals. B) Total average duplication and deletion lengths comparison between AU and TD individuals. Page 27

33 Table 5. A list of rare, potentially pathogenic events that were removed from the HMM calls for non-pathogenic CNV burden analysis. These events were found in 10 or less of 8329 controls. Some of these variants have been previously described. Chr Section CNV Start End Size Samples ID Controls chr4 4q32.3 Deletion /8329 chr15 15q11.2q12 Duplication /8329 chr15 15q11.2q12 Duplication /8329 chr15 15q11.2q12 Duplication /8329 chr15 15q13.3 Duplication /8329 chr15 15q13.3 Duplication /8329 chr7 7q11.23 Deletion /8329 chr17 17q12 Duplication /8329 chr16 16p11.2 Deletion /8329 chr16 16p11.2 Duplication /8329 chr16 16p11.2 Duplication /8329 chr1 1q21.1 Deletion /8329 chr17 17p11.2 Deletion /8329 chr17 17p11.2 Deletion /8329 chr4 4q13.1 Duplication /8329 chr7 7p12.3 Duplication /8329 chr10 10q11.23 Deletion /8329 chr7 7q11.22 Deletion /8329 chr6 6q22.31 Duplication /8329 chr6 6q23.2 Deletion /8329 chr6 6q26 Duplication /8329 chr12 12p11.1 Duplication /8329 chr22 22q11.22 Duplication /8329 chr15 15q13.1 Duplication /8329 chr17 17q12 Deletion /8329 chr22 22q11.21 Duplication /8329 Page 28

34 Large events constitute the elevated level of duplication When we separated each of the CNVs into several different classes based on their lengths to assess the frequency distribution of the CNV events in each cohort, interestingly, we found that the elevated level burden of duplications we observed earlier in the autistic individuals was mostly due to the large size events, with lengths of CNVs more than 200 kb (Figure 9B). This suggested that the elevated level of duplications in AU individuals compared to that of controls was principally contributed by large duplicated segments of CNVs. On the other hand, there was no such observable trend in the deletion events, which corresponded to our expectations since we found no significant difference in the burden level of deletion between cases and controls. The test for the significance of the data was not performed since the frequency plot was graphed mainly for the purpose of observing the trends on what types of CNVs (in terms of size) contributed to the elevated level of burden we observed earlier. Page 29

35 Figure 9. Frequency distribution plots of CNV events in both cohorts, separated into different classes based on sizes. A) Deletions: There is no observable trend in the distribution of deletion events in the comparison between AU and TD individuals. B) Duplications: The duplications events found in AU individuals mostly fall into the large CNV classes ( 200 kbp), leading to the elevated level of duplication we observed earlier. Page 30

36 DISCUSSION In the design of our Hotspot v1.0 array, the segmental duplication architecture of the human genome [26] was utilized to custom-design a DNA oligonucleotide microarray enriched for genomic hotspots, which are regions surrounded by high-identity level segmental duplications. Because the hotspot regions were the main target with higher probe density designed to cover those regions, this array actually had more powerful detection sensitivity within hotspot and hotspot-associated regions compared to several other available commercial arrays. In fact, recurrent events detected in the genomic hotspots were twenty-five times more frequent comparing to the rest of the genomic backbone regions [27]. Many studies have been focused on discovering the large, rare pathogenic CNVs that lead to a variety of neurodevelopmental diseases such as autism, mental retardation, schizophrenia, intellectual disability as well as Prader-Willi/Angelman syndrome, in which those structural variations often involved copy number changes in genomic segments that harbor large functional cluster of genes, leading to the pathogenesis of these diseases [28, 29]. However, some other studies focused on a disease association with copy number polymorphism, which included the events that were much more common in the population than rare pathogenic CNVs [30]. However, very few of these studies have actually evaluated the relationship between global CNV burden (not limited to rare, genic CNVs) and pathogenesis of a particular disease. Therefore, the question whether there is a higher level of genomic instability in autism has been left unanswered for decades. In this study, we were able to at least partially address that question by demonstrating the increased level of genomic instability in autistic individuals. Page 31

37 Autistic individuals have higher level of global CNV burden Our array and statistical results showed that there was a significantly higher level of overall global CNV burden in autistic cases compared to that of the controls. This result is in agreement with a previous study where the authors reported an increased level of copy number burden in AU individuals using the whole genome studies focused on large de novo events [31]. Another study of genome-wide assessment via single-nucleotide polymorphism microarrays and karyotyping has also found that structural variants or unbalanced CNVs were present in relatively higher frequency in individuals with autism spectrum disorders (ASD) [34]. In a more recently published work, an elevated level of copy number burden was detected for rare pathogenic variants primarily associated with genes previously implicated in ASD or intellectual disability (ID) [32]. In addition to that, numerous studies have also discovered high level of genomic imbalance in other diseases or psychiatric disorders such as schizophrenia [35-37], bipolar disorder [38], epilepsy [39,40], as well as mental retardation [41] and developmental delay/intellectual disability [42,43]. However, these studies only focused on examining the burden level of de novo events of pathogenic or functional CNVs [33], but did not include those common CNVs or events that are not pathogenic (not harboring genes that might influence disease susceptibility), and have not investigated whether the increased instability occurred in hotspot or non-hotspot genomic regions. Nevertheless, these findings are in agreement with our principal result that there is indeed an increased level of copy number burden in autistic individuals compared to unaffected controls. The increase of global CNV burden in autism demonstrates the important role of genomic structural variants in autism. Page 32

38 Increased level of CNV burden in autism is manifested by large duplications We have also shown that the elevated level of CNV burden was mainly represented by large duplications ranging from 200kb to 5Mb based on the frequency distribution of CNV in different size classes. This is in agreement with many other studies that have reported large duplicated CNVs such as those located in chromosomal segments associated with autism: 7q11.23 [60], 15q [61,62], as well as 16p11.2 [63,64]. We speculate that from an evolutionary point of view a large duplication of a chromosomal segment was relatively more tolerable compared to a deletion in the human genome, since deletions totally disrupt and eliminate some functional genes that might be critical and essential for the survival of the organism. Hence, large duplications in the genome provide a higher chance for an organism to survive, but at the same time, could affect some functional genes dosages leading to certain observable disease phenotype such as autism. Remarkably, the increased level of CNV burden in AU individuals was still statistically significant even after the rare, pathogenic CNVs were excluded from our analysis. This indicates that the increased genomic instability we have observed in affected individuals is not exclusively caused by the rare, pathogenic CNVs. This could serve as a starting point for a future research to confirm that genomic instability and copy number burden in certain disorders, at least in autism, may not be limited to the rare and large genic events. Page 33

39 Copy number burden is elevated in non-hotspot genomic regions When we performed the CNV burden level analysis separately on the genomic hotspot and nonhotspot regions, unexpectedly, we found that the elevated level of copy number burden was mainly localized in the non-hotspot regions. This finding question the current paradigm that assumes that majority of the specific genetic diseases are caused by homologous recombination and recurrent chromosomal aberrations, influenced by the presence of segmental duplications [48-50]. Hence, it is certainly reasonable to think that mechanisms other than non-allelic homologous recombination (NAHR) could be involved in the elevated level of non-hotspot CNV burden we observed in the AU individuals. One of the possible factors that contributes to genomic instability may be the presence of Alu elements, which were considered crucial for chromosomal rearrangements and aberrations in the human genome in the last two decades [51,52], and have been thought to mediate genomic instability in diseases such as breast cancer [53,54], colon cancer [55], leukemia [56], and many other human diseases [57]. In fact, a recently published study [58] has shown that a disruption of an important neuronal gene associated with ASD, CNTN4 [58,59], was a result of Alu Y-mediated unequal recombination. It would be certainly reasonable to consider other mechanisms not mediated by segmental duplications, including microhomology-mediated break induced replication that could potentially lead to a higher genomic instability and an increased level of CNV burden reported by this and other studies. Page 34

40 Genomic instability, copy number variations and autism One interesting implication of our result is that an increased level of copy number changes could result from a large number and variety of genetic alterations, only part of which may be involved in pathogenesis of the disorder. It is possible that some specific external or internal factors may exist that promotes this higher overall instability. For instance, environmental factors that affect genomic stability could play a role, contributing to the pathogenesis of autism. The idea of connection between environmental factors and genomic instability is certainly not new; many studies reported the defined and clear relationship between these two factors for many other diseases including cancer [44] and many complex genetic disorders [45,46]. Hence, it would certainly be interesting to investigate how various environmental factors affect the genome, which in turn may lead to genomic structural variants and ultimately the observable disease phenotype. As previously mentioned in many other studies and literature reviews, even though most common CNVs do not necessarily have a negative effect on health (normally, CNVs cover approximately 12% of the human genome [48]), when their amount reaches a certain threshold, they could have a number of important implications in the causal mechanism of genetic disorders. The association of neurological disorders such as autism with the increased CNV burden suggests an important role of the non-mendelian inheritance in the emergence of the disease. An individual's genetic code may not be simply the addition of the genetic contributions from both of the individual s parents. The mechanisms such as unequal crossover events, which occur during the production of sperm and eggs, are responsible for formation of novel CNVs in the Page 35

41 genome. As a result, the progeny may lose or gain additional copies of genetic information that were not present in either of their parents' genetic code. For many years, geneticists believed that a disease is the result of inherited genetic variants from the previous generations, where a malfunctional copy of a genomic segment was passed on to the next generation. However, as more and more studies started to discover and recognize other mechanisms that can lead to genetic disorders, we are in an era where we begin to understand and appreciate that structural variants may provide the genetic basis for increasing the risk of common and complex diseases such as mental retardation, schizophrenia, and autism. Limitations of the Hotspot v1.0 array The Hotspot v1.0 array certainly has its limitations in terms of detecting CNVs in the human genome, due to the differential probe density across the genomic hotspot and non-hotspot regions. As mentioned earlier, the array was originally designed to target genomic hotspot regions flanked by segmental duplications by increased density of probes in these regions (median probe spacing 2.6 kb) compared to the genomic backbone. Consequently, the array design was biased towards targeting hotspot regions in the genome, and has its limitations in detecting CNVs that are located in the non-hotspot regions. Specifically, it was not possible to locate smaller and shorter CNVs due to the low density of probes in those regions (the minimum probe spacing about 20 kb), and the requirement for a CNV to contain at least 10 probes. Hence, the non-hotspot CNVs that were smaller than 200 kb completely missed our detection in the nonhotspot regions. In other words, we would inadvertently overlook those smaller and intragenic as well as exonic CNVs in the genomic backbone that might also influence the expression of other autism candidate genes. Page 36